Google has updated its Gemini app for Mac so the AI assistant can work across applications on a user’s computer, carrying out multi-step tasks rather than limiting its help to a chat window. The feature is launching in the United States and requires a Google AI subscription.
From prompts to multi-step workflows
To start a task, users include the command “@Computer use” in a prompt. Gemini can then identify a suitable local application, outline a plan and execute the work step by step. Google says a workflow may involve dozens of actions, making the feature more suited to extended tasks than to a single request for information.
One example is a designer providing a vehicle sketch and asking Gemini to turn it into a three-dimensional animation. The assistant can select the needed software and proceed through the work. The example illustrates a shift from generating a response inside a conversation to operating tools on the user’s behalf; the particular applications and results will depend on the task and the apps available on the computer.
Workflows can also continue from one application to another. A user might ask Gemini to produce a visual asset in a design program and then use that output as the starting point for a project in Visual Studio Code. This ability to hand off results could reduce the effort involved in moving between stages of a task, although the assistant’s role remains bounded by the applications and permissions the user makes available.
Local apps, web services and user oversight
The Mac app’s computer-use feature is not limited to software installed on the device. Gemini can also interact with web services, including bringing information from webpages into local programs and transferring data in the other direction or between cloud services. That gives the assistant a way to connect work that would otherwise require a user to move information manually between browser tabs, desktop applications and online tools.
Gemini runs its work in a background window, so users can continue other activities while a workflow is underway. A preview pane in the Mac client’s chat window lets them follow the process. Google has also added controls for feedback during a task. Michael Friedman, Google’s group product manager for the Gemini app, said users choose which apps Gemini can access, can watch its steps, pause the workflow and approve key actions before they happen.
Those controls are significant because an assistant that can operate software has a different role from one that only drafts text or answers questions. It may handle sequences of actions and move information across services, so users need a way to monitor what is happening and intervene. Google’s description emphasizes permission and approval, but the company’s announcement does not establish that every action in every app requires the same level of confirmation.
A competitive step, with business expansion still open
The launch also places Google in a growing contest over AI agents that can use computer applications. Meta Platforms’ Mac version of Muse can interact with local apps and data, and its cloud-based virtual machine can continue tasks when a user’s computer is offline. OpenAI’s Dots offers a similar computer-use capability and includes automation features aimed at business workflows such as procurement.
Google’s release is currently limited to the Mac app in the United States and requires a paid Google AI subscription, which also includes additional storage and other features. The company could extend computer use to other operating systems or add tools for business customers, but those are possible future directions rather than features announced as available in this release. For now, the update gives subscribers a way to delegate cross-app work while retaining visibility and control, and leaves broader platform availability and business-specific automation unresolved.